|Veröffentlichungsdatum||9. Febr. 2010|
|Eingetragen||24. Sept. 2003|
|Prioritätsdatum||24. Sept. 2002|
|Auch veröffentlicht unter||US20040215726|
|Veröffentlichungsnummer||10670638, 670638, US 7660855 B2, US 7660855B2, US-B2-7660855, US7660855 B2, US7660855B2|
|Erfinder||Andreas Arning, Roland Seiffert|
|Ursprünglich Bevollmächtigter||International Business Machines Corporation|
|Zitat exportieren||BiBTeX, EndNote, RefMan|
|Patentzitate (4), Referenziert von (18), Klassifizierungen (17), Juristische Ereignisse (5)|
|Externe Links: USPTO, USPTO-Zuordnung, Espacenet|
1. Field of the Invention
The present invention relates to the field of computer technology, and in particular to a computerized method for predicting the addressee field in an electronic mail system, in which history information associated with the user's sent mail is analyzed for the purpose of associating the most probable addressee for a given e-mail letter.
2. Description of the Related Art
When using electronic mail (e-mail), a user normally must use the addressee's complete electronic address, referred to herein as a TCP-IP address. The user is typically required to use the TCP-IP address in a correct form, because this address is used to route the mail to the intended target mail server. Usually, there is an IP-address of the mail server encoded within the TCP-IP address. The TCP-IP address further includes user-specific ID-information (e.g., the name or nickname of the addressee) so as to distinguish one particular addressee from among a plurality of users registered with the target mail server.
Usually, the TCP-IP-address information itself is not actually entered by the user, but instead, a more readable character string is used. For example, the character string that comprises the TCP-IP address can be converted by the personal e-mail client program into the complete name of a respective person, e.g., “David Miller” might be used and displayed to the user instead of the TCP-IP address “Miller-David@companyX.de”.
This additional information concerning location, organization and/or country is usually added to the user name to make it unique. Sometimes, even a number may be added to the name to generate a unique mail system address.
State of the art e-mail client programs installed at the user's PC do not propose or predict an addressee term, i.e. the addressee's e-mail address, when a new mail is created. Once, however, the user begins to enter some characters into the addressee field in the ‘new message’ window of his e-mail client, some programs will attempt to complete the name based on this initial entry of characters. This is true, for example, for newer versions of Microsoft™ Outlook Express™. A similar procedure is used in Microsoft™ Internet Explorer, Version 6 and many other programs, which append the rest of a stored character string to the initial string entered by a user. A disadvantage of this automatic string-completion is that it requires an initial user action to identify a significant subset of addressees, which are selected from a personal address book stored locally at the user PC, and/or stored in a centralized form at the e-mail server in an intranet or LAN of an enterprise.
If the user creating a new e-mail does not enter anything into the addressee field, prior art techniques disadvantageously do not make any proposal with regard to whom the mail could be addressed, even if the header/subject line is already filled-in by the user, and/or the text of the mail already comprises a considerable number of pages.
If some initial entry, e.g., at least one character, is present in the addressee field, a prior art e-mail program, for example the e-mail client in Lotus Notes™, Release 5, performs the above-mentioned address prediction based on stored addresses at the mail server of an Intranet, but to do that, the user's PC must be connected to the LAN, or to the Intranet, respectively. If the user's PC is cut off from such network, only the locally stored address book may be used for addressee term proposals.
Thus, where access to the complete list of electronic mail addressees is available, i.e., in a “closed world” scenario, where basically all mail addresses are known and can be listed or accessed, a prior art e-mail client may do one of the following solutions when confronted with an incomplete or ambiguous mail address entry:
Such predictive methods, however, result in unsatisfactory proposals which, in many cases, are nonsensical and may result in misrouted electronic mail.
It is thus an objective of the present invention to improve the prediction accuracy of auto-fill email addressing.
The present invention combines prior art Text Mining methods, such as those methods commercially available in the IBM™ product “DB2 Intelligent Miner for Text”, and prior art Data Mining methods, such as those methods commercially available in the IBM™ product “DB2—Intelligent Mining for Data” (both products being available at each IBM business partner and publicly available at http://www.ibm.com/software/data/iminer/) to automatically determine the most likely addressee in a new e-mail drafted by a user. Prior art Text Mining methods are applied to a set of attributes which are characteristic for e-mail correspondence in order to generate intermediate results in a particular form, preferably in a table-based form, which may be further processed by prior art Data Mining methods. Model generation and training is done once, and then repeated at predetermined intervals, or when it seems appropriate. The application of the generated and trained models is preferably done on-the-fly and transparent to the user, whenever the entry of an addressee is needed. Further training may occur at a place different than the location of the application, and may be undertaken by different enterprises.
According to one aspect of the present invention (a training mode), a computerized method is disclosed for predicting the correct addressee to be filled-in in an addressee field in a personal electronic mail system, by which user-related history information, including the user's sent and/or received mail, is analyzed for associating the most probable addressee for a given e-mail letter. The inventive method of the embodiment is characterized by the steps of:
a) analyzing the contents of at least a subset of the following attributes for both the newly-composed mail and historical mail:
b) weighting the plurality of single-analysis results to provide a Data Mining Model adapted to offer at least one top favorite addressee proposal as a prediction result.
It should be noted that when using the received-mail history information, the information from the sender field should be used instead of the addressee field (which is used when using the sent-mail history information). Of course, each information source may be used separately, or they may be used in combination.
Further advantageous embodiments and improvements will be apparent from the descriptions and appended claims.
The present invention is illustrated by way of example and is not limited by the illustrations of the drawings in which:
With general reference to the figures and with specific reference now to
At step 110 an inventive program module, which may be incorporated in a personal e-mail client local to the user, or, alternatively in a centralized form at the mail server/firewall, accesses the sent mail and the received mail of a particular user. This “history data” is now subjected, according to the present invention, to prior art Text Mining methodology, at step 120. Specifically, a major subset of the history data is used as a training set for predicting the one or more addressees of a mail by exploiting the features, which are specific for e-mail applications, such as the following attributes:
The subject line, the salutation and closing sequences, the language or vocabulary in use, and the topics, which are discussed in the body of each e-mail, and optionally any time information available, either in the plain text of the body, or available as meta data in the e-mail client program, and optionally the lengths of the mail. The product of this text mining process is stored as intermediate results (step 130).
The structure of data, which is used for analyzing the above-mentioned attributes (subject line, etc.) may be selected in any manner, but preferably in a form which maximizes its efficient use in the succeeding step (step 140) of processing by a Data Mining method. A preferred data structure to accomplish this is illustrated by way of example in
According to a basic principle of the present invention and of the preferred embodiment, this further analysis at step 140 comprises the use of known Data Mining based methods. The result thereof is a trained prediction model (step 150), which is able to associate a given new e-mail to be created with a small set of potential addressees. The resulting association forms a list of “top candidates” (e.g., one or two top candidates) that have a high probability of being the correct e-mail address desired by the author of the e-mail. The list may be generated by the inventive program module and may be issued to the user for selecting the correct addressee, for instance in case three or four addressees are proposed.
In a further, optional step the trained models created at step 150 may be subjected to a test phase (step 155), as this is a common procedure in prior art model build-and-test arrangements. A separate test set of history data, which is preferably unique (has no shared elements) to the training set, is used to predict the addressee, and the system's prediction is compared with the actual addressee. If the rate of correctly predicted addressees is high enough, for instance a success rate of 90%, the trained models created at step 150 may be used for prediction to be applied for newly created e-mails; thus, some prediction result will be obtained (step 170), whenever a user starts writing the plain text of a new e-mail letter, even if the user did not fill in any character string into the addressee field.
Once the model is created and optionally tested during the test step 155, the model 150 may be used in its application mode to do one or more of the following:
A first task is to automatically propose an addressee for an e-mail draft (step 160), which is already written and in which the addressee field is still empty. Thus, an e-mail draft (step 160) is applied to the prediction model (step 150), resulting in a prediction result (step 170). This may cause the addressee field to be pre-filled. If the prediction method in use is able to return a confidence value for a prediction, this confidence value may be compared to a threshold and thus help to suppress insignificant predictions. Such threshold levels may be predetermined to comprise several different values dependent upon their use. For example, in private use the thresholds may be lower and in business use they may be higher in order to reflect a higher level of scrutiny which may be necessary in business practice.
Further, the prediction result achieved at step 170 may be used to automatically expand ambiguous addressee information according to the best match. This is done preferably by giving a “highest priority” designation to the candidate having the highest confidence value. In an online mode in the case of a “closed world” scenario, as described above, this addressee term expansion may be visible as a list, from which the user may select a name. In an offline mode, the result achieved at step 170 may also (or instead) be used to improve the current address resolution to avoid sending an e-mail to a non-intended addressee.
Further, the prediction result achieved at step 170 may be used for a cross-check between the proposed addressee and the addressee entered manually by the user, in order to raise, i.e. issue a warning, if a predicted addressee is not on the addressee list, and/or if a not-predicted addressee is on the list. If the prediction method used is able to return a confidence value, this value can be compared to appropriately selected threshold values, in order to determine, whether either of the above warnings (1. predicted, but not on addressee-list, 2. on addressee-list, but not predicted) should be raised.
With reference now to
In a preferred embodiment, the following attributes are evaluated according to the present invention:
The subject line showing some header topic 210, which gives the reference (RE) information, may be analysed by Text Mining, to retrieve similarities between different topics present in the training data.
Further, the salutation and/or closing attributes 215 may be analyzed, for example, to determine the degree to which they are more, or less, formal. The salutation “Hello Joe” could be associated as a formal salutation for instance, with a confidence value of only 5%, whereas a salutation “Dear Sirs” could be assessed as a formal salutation with a confidence value of nearly 100%. Similar rules could apply for intermediate degrees, e.g., for cases such as “Hello Mr. Miller” etc.
In the same way the language style attributes (language formality 220), i.e., the vocabulary used in the body of the e-mail, may be analyzed as being more or less formal. Slang-type wordings could contribute to a language classification of “informal”, whereas the absence of such slang could generate a confidence value of high degree that the language of the e-mail corresponds to a formal style of language.
Further, in accordance with the present invention, the language may be analyzed by Text Mining methods, in order to reveal a confidence value for a predetermined selection of languages. In
According to the invention, a further attribute for evaluation is the confidentiality 240 of an e-mail. According to the invention, an existing e-mail may be identified as confidential, if one or more of predetermined keywords (e.g., “confidential”) exist either in the subject line, or in the body of the text. Further, other meta information may be used to assess a mail as confidential. For example, if the mail has an attachment in an encoded form, the message can be identified as being confidential.
Further, topics 25 [250?] in the text body may be associated with one or more predetermined categories, examples of which are shown in
Each of those topics is again accompanied by a percentage field which is provided to store the confidence value calculated by the applied text mining method. Thus, a given e-mail may achieve a score of 30% politics, 20% sports and 50% sale of product x (which corresponds to project 1). The Text Mining techniques used to perform this analysis corresponds to prior art methods.
The method of the present invention may be implemented in a way which provides one prediction model per user, i.e., end model is unique to each user. This approach may, of course, be varied because there may be cases where a group of users may wish to share the same prediction model, as they share the same contacts. This helps to simplify and standardize the prediction. Further, the method of the present invention may be implemented to cover the usually prevailing three different types of addressee lists, as are “to”, “cc:”, and “bcc”. It may be used to either list separately, and/or in common, lists of likely addresses to be used in these fields.
When using separate Data Mining models for different use modes, e.g., office/private mode of user activities, German/English language used in the mail, Confidential/Not Confidential mail, etc., the underlying models may be selectively trained and prediction quality may be increased.
The same advantage may be obtained when performing a retraining of the user-specific Data Mining model triggered by either of the following criteria:
a) when a user overwrites the addressee proposals made by the e-mail system, more frequently than limited by a predefined threshold level,
b) when the e-mail system is confronted with a number of new addressees, which do not form part of the user's history, and the number or fraction thereof is higher than a predefined threshold level,
c) after a predefined time limit has passed.
Thus, the underlying prediction model is always up-to-date.
When said analysis results are generated in a table-like form, in which each attribute to be analyzed is associated to its predicted value, accompanied by a respective confidence value, then a preferable output form is obtained, which is best worked on subsequently by prior art Data Mining tools, if they are used within the inventive method.
According to another aspect of the present invention (an application mode), a computerized method for completing the addressee field in a user-initiated “new mail” within an electronic mail system is disclosed, which is characterized by the steps of:
a) on occurrence of an incomplete entering of the addressee term in an addressee field, running a predictive Data Mining method based on the trained Data Mining Model as mentioned above,
b) determining at least the most probable addressee to the user as a prediction result.
Another embodiment comprises the further step of offering a subset of a predefined quantity of top favorite addressee proposals to the user, and a respective plurality of similarly assessed top favorites may be proposed to the user to allow selection of the correct single address, or to select further addressees for cc—(copy) or bcc—(blind copy) lists.
The present invention functions particularly well when automatically providing an addressee field pre-filled with the top favorite addressee, where a single proposed favorite has a significantly higher ranking result than the other proposed favorites.
When testing the Data Mining model on a test set of mails, not being part of the training step, before predicting the most probable addressee, and issuing a hint to the user, indicating the confidence of the predicted addressee proposal, for instance in form of a confidence value, the user is given additional security for the correct addressee selection, which may save time otherwise needed for a confirmatory check of the personal or Intranet-wide address book.
When further, in case of a trunk of the addressee term being present in the addressee field, e.g., due to manual entering, and in case a high significance of the predictable addressee being present provided by the run of the Data Mining method, said trunk may be automatically expanded with the most probable addressee term. This also contributes to increased user comfort.
When finally, cross-checking a term entered manually by the user with a list of top favorite addressees, determined by the inventive system, and issuing a warning, if the probability is high that the user-entered term is faulty, then an additional contribution is provided to avoid misrouted mails.
In addition to being useable for determining the most probable addressee of a newly-drafted e-mail, the concepts of the present invention can also be applied for determining, for example, if such e-mail letter should be sent as “CONFIDENTIAL INFORMATION” or not, thus implying the need to use appropriate encoding techniques, or not. For this purpose, a training phase can be run to create a model which helps to decide if a document is confidential or not. For instance, such training may generate a predetermined list of keywords, and the letter draft is scanned for such words, for example during the Text Mining processing of “topics”, as described above. If one or more of such keywords occur, the user can be alerted that the mail should be sent according to the rules usually applied for confidential information.
The present invention can be realized in hardware, software, or a combination of hardware and software. A prediction tool according to the present invention can be realized in a centralized fashion in one computer system or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods described herein is suited. A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein.
The present invention can also be embedded in a computer program product, which comprises all the features enabling the implementation of the methods described herein, and which—when loaded in a computer system—is able to carry out these methods.
Computer program means or computer program in the present context mean any expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following
a) conversion to another language, code or notation;
b) reproduction in a different material form.
|US6005928 *||15. Apr. 1997||21. Dez. 1999||Mci Communication Corporation||Method and system for automatic distribution addressing|
|US6684238 *||21. Apr. 2000||27. Jan. 2004||International Business Machines Corporation||Method, system, and program for warning an email message sender that the intended recipient's mailbox is unattended|
|US20020165932 *||13. Apr. 2001||7. Nov. 2002||Rensin David K.||Systems and methods for automatically accessing internet information from a local application on a handheld internet appliance|
|US20030115279 *||14. Dez. 2001||19. Juni 2003||Pitney Bowes Incorporated||System and method for address correction of electronic messages|
|Zitiert von Patent||Eingetragen||Veröffentlichungsdatum||Antragsteller||Titel|
|US8543404 *||7. Apr. 2008||24. Sept. 2013||Nuance Communications, Inc.||Proactive completion of input fields for automated voice enablement of a web page|
|US8595230 *||27. Juni 2012||26. Nov. 2013||Google Inc.||Intelligent identification of form field elements|
|US8825654 *||25. Okt. 2012||2. Sept. 2014||Netseer, Inc.||Methods and apparatus for distributed community finding|
|US8831950||7. Apr. 2008||9. Sept. 2014||Nuance Communications, Inc.||Automated voice enablement of a web page|
|US8838605||25. Okt. 2012||16. Sept. 2014||Netseer, Inc.||Methods and apparatus for distributed community finding|
|US8843434||28. Febr. 2007||23. Sept. 2014||Netseer, Inc.||Methods and apparatus for visualizing, managing, monetizing, and personalizing knowledge search results on a user interface|
|US9047869||7. Apr. 2008||2. Juni 2015||Nuance Communications, Inc.||Free form input field support for automated voice enablement of a web page|
|US9110985||15. Okt. 2010||18. Aug. 2015||Neetseer, Inc.||Generating a conceptual association graph from large-scale loosely-grouped content|
|US9443018||12. Aug. 2014||13. Sept. 2016||Netseer, Inc.||Systems and methods for creating, navigating, and searching informational web neighborhoods|
|US9686219||14. Apr. 2010||20. Juni 2017||Nokia Technologies Oy||Systems, methods, and apparatuses for facilitating determination of a message recipient|
|US20070203903 *||28. Febr. 2007||30. Aug. 2007||Ilial, Inc.||Methods and apparatus for visualizing, managing, monetizing, and personalizing knowledge search results on a user interface|
|US20080104061 *||24. Okt. 2007||1. Mai 2008||Netseer, Inc.||Methods and apparatus for matching relevant content to user intention|
|US20090254346 *||7. Apr. 2008||8. Okt. 2009||International Business Machines Corporation||Automated voice enablement of a web page|
|US20090254347 *||7. Apr. 2008||8. Okt. 2009||International Business Machines Corporation||Proactive completion of input fields for automated voice enablement of a web page|
|US20090254348 *||7. Apr. 2008||8. Okt. 2009||International Business Machines Corporation||Free form input field support for automated voice enablement of a web page|
|US20090300009 *||1. Juni 2009||3. Dez. 2009||Netseer, Inc.||Behavioral Targeting For Tracking, Aggregating, And Predicting Online Behavior|
|US20110113032 *||15. Okt. 2010||12. Mai 2011||Riccardo Boscolo||Generating a conceptual association graph from large-scale loosely-grouped content|
|US20130046842 *||25. Okt. 2012||21. Febr. 2013||Netseer, Inc.||Methods and apparatus for distributed community finding|
|US-Klassifikation||709/206, 704/10, 704/3, 704/8, 709/225, 709/207, 709/203, 709/227, 707/758|
|Internationale Klassifikation||G06F15/16, G06F17/27, G06F17/21, H04L12/58|
|Europäische Klassifikation||H04L12/58, G06F17/27P|
|22. Juni 2004||AS||Assignment|
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:ARNING, ANDREAS;SEIFFERT, ROLAND;REEL/FRAME:014760/0815
Effective date: 20040524
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION,NEW YO
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:ARNING, ANDREAS;SEIFFERT, ROLAND;REEL/FRAME:014760/0815
Effective date: 20040524
|30. Nov. 2010||CC||Certificate of correction|
|20. Sept. 2013||REMI||Maintenance fee reminder mailed|
|9. Febr. 2014||LAPS||Lapse for failure to pay maintenance fees|
|1. Apr. 2014||FP||Expired due to failure to pay maintenance fee|
Effective date: 20140209